Inside Google’s Targeting Plans for a Post-Cookie World

Latest ad targeting experiments in Chrome show how ad tech is shifting toward a decentralized system

The Google logo with darts in it
Cohort, or FLOC, targeting forms part of Google Chrome's Privacy Sandbox initiative. Getty Images, Google
Headshot of Ronan Shields

Earlier this week, the Justice Department filed suit against Google citing its dominance of the search market as anti-competitive. But away from the theater of big government versus big tech, advertisers are paying attention to the countdown to the “cookie-pocalypse” as Google prepares to withdraw support for third-party cookies in its Chrome browser.

This, of course, is prompted by changes in user tracking laws across the world, and after making this monumental announcement in January, Google’s Chrome team has been quietly rolling out its Privacy Sandbox initiative.

This project represents Google’s attempts to figure out how online advertising will work after the devaluation of the industry’s universal currency. And while not all of its proposals have been met with acclaim, the project is arguably the most realistic portal of how the $125 billion industry will operate post-2022.

This week, Google published the results of recent experiments that took place as part of the Sandbox initiative with the analysis centered on the concept of replacing one-one targeting—the central premise of cookies—with cohort or “FLoC” targeting to study the habits of similar user groups.

“More precisely, the FLoC API relies on a cohort assignment algorithm,” according to the report. “That is a function that allocates a cohort ID to a user based on their browsing history.”

Other fundamentals to FLoC targeting include:

  • A cohort ID should prevent individual cross-site tracking
  • A cohort should consist of users with a similar browsing history
  • A user’s browsing history should be “hashed” to prevent identification

The precision of the algorithm’s targeting capabilities were tested against publicly available datasets that would cluster certain interests or verticals.

The online ad giant claimed this process of audience targeting improved 350%, and observed 70% improvement in targeting precision “at very high anonymity levels, compared to random grouping.”

The authors of the report acknowledge some of the “privacy-utility trade-off” in the system. For instance, the larger the cohort, the better assurance of privacy. Although, the larger the cohort, the more difficult it is to serve a user with a relevant ad.

“Nevertheless, it is promising to see that a fully decentralized approach can generate cohorts achieving approximately 85% of the quality of a fully centralized algorithm,” the report said.

A team effort?

Google is encouraging other ad-tech companies to perform similar experiments testing its algorithms on their own proprietary datasets.

Additionally, it is taking feedback from third parties, primarily via the web standards body W3C, where peers are usually not shy of voicing dissent, even if Google’s detractors in Congress assert that it has an outsize influence there.

Some claim Google’s findings have its own interests at heart, despite it inviting third-party feedback. Additionally, privacy advocates have voiced their problems with such new approaches to online ad targeting.

@ronan_shields Ronan Shields is a programmatic reporter at Adweek, focusing on ad-tech.